ww

w.in

l.gov

RAVEN/RELAP5-3D UQ analysis for a DOE LWRS/RISMC Multiple Hazards Demonstration

Carlo Parisi, Steven R. Prescott, Justin L. Coleman, Ronaldo H. Szilard..& the RAVEN team

NEKVAC/NUC Workshop “Multi-physics Model Validation”June 27-29, 2017The James B. Hunt Jr. LibraryCentennial Campus – NCSU, Raleigh, NC, USA

Topics

• LWRS/RISMC Multiple (External) Hazards Demo Motivation

• Toolkits

• Main steps

• Testing RELAP5-3D/RAVEN capabilities

• Limit Surface Search

• UQ for External Events

2

Motivation• Goals of the RISMC Pathway

– Develop and demonstrate a risk-assessment method coupled to safety margin quantification

– Create an advanced “RISMC toolkit”

• Motivation: to perform an advanced risk analysis of accidental events caused by a combination of natural external hazards, i.e. earthquake and flooding

- Use of INL advanced simulation tools & methods

- Perform realistic risk analysis for a generic PWR/BWR

- Study NPP behavior under:

• Internal/External flooding scenario caused by EQ (e.g., EQ-induced pipe rupture, levee break)

- Outcomes for FY16:

• Risk analysis of scenarios caused by external events, using realistic plant models, simulations and uncertainties (for a generic PWR)

- Two toolkits (External EVEnts toolkits, EEVE) + pathways are defined

3

Traditional vs. Advanced Method

Advanced External Hazards analysis workflow

4

• Advanced Component [EEVE-A]

• Use of advanced INL tools (e.g., RELAP-7),

• Direct coupling (e.g., flooding-RELAP7),

• Use of Reduced Order Model (ROM) & Surrogates

• Advanced Seismic probabilistic risk analysis (ASPRA)

The Toolkits: EEVE-B & EEVE-A• Baseline Demonstration [EEVE-B]

• Use of existing, validated & state-of-the-art tools (e.g., RELAP5-3D)

• one-way coupling

• generic NPP

5

Industry Application #2 - Methodology

5 Phases identified for the EEVE-Baseline– Simulate effects of EQ on a NPP SSCs using

advanced seismic analysis methodology• Use of Non-linear soil-structure interaction (NLSSI)

methodology [LS-DYNA & MASTODON]• Piping fragilities evaluation [OPENSEES]

– Simulate NPP flooding scenarios caused by EQ-induced pipe rupture [NEUTRINO]

– Simulate NPP primary circuit + part of BOP dynamics [RELAP5-3D]

– Apply S/U analysis [RAVEN]– Evaluate risk of scenarios (ranking) using dynamic

PRA analysis [EMRALD]

0 10 20

5−

0

5

MASTODON

LS-DYNA

NEUTRINO

EMRALD

RELAP5-3DRAVEN

IGPWR Basic Information

• INL-Generic PWR (IGPWR) defined for EE analysis

• Main Characteristics:– 3 Loop PWR / NSSS by Westinghouse

– Core average power: 2546 MWth [855 MWe]

– Core: 157 FA [15x15 Westinghouse FA]

– Sub-atmospheric Containment

6IGPWR ESF

Seismic Analysis

7

• Calculation of Non Linear Soil-Structure Interaction (NLSSI) by LS-DYNA/MASTODON code

– Use of generic soil

– Propagation of EQ ground motion

– Acceleration Response Spectra

• Piping analysis by OPENSEEScode

– Determination of fragility curves (PGA vs Probability of Failure)

Seismic Hazard CureNLSSI Analysis

Acceleration Response Spectra for Aux Building

Structural Analysis of Fire Suppression System Seismic Fragility Curves

PRA Analysis – SAPHIRE/EMRALD codes

• PRA model – Updated SAPHIRE model for generic 4-loops PWR to generic 3-loops PWR– Added External events/EQ ref. “NRC Risk Assessment of Operational

Events Handbook Volume 2 – External Events” – PWR Simplified LOOP and SBO PRA (event trees + fault trees) developed– Convert SAPHIRE PRA to EMRALD PRA models– EMRALD calculates PRA dynamically:

• Run RELAP5-3D/RAVEN [when possible fuel damage could occur] & NEUTRINO[when flooding could occur] codes

• Process RELAP5-3D/RAVEN & NEUTRINO results into the final result probabilities

8

IA#2 Specific Fault Tree

Loss-of-Offsite Power Event Tree

Station Blackout Event Tree

IA2-ACP-1AA02

4160 VAC BUS 1AA02 FAULT

IA2-EPS-RISMC22

FAILURE OF DG-A DURING A LOOP

ExtIA2-EPS-DG1A-RISMC

DIESEL GENERATOR 1A FAULT TREE FOR RISMC IA #2

IA2-ACP-CRB-AA20

SWITCHYARD AC BREAKER AA20

2.500E-03ACP-CRB-CC-AA20

FAILURE OF SWITCHYARD AC BREAKER AA20 TO OPEN

4.250E-05ACP-CRB-CF-A2030

CCF OF SWITCHYARD AC BREAKERS TO OPEN

AA0205/0301

IA2-EPS-SEQ-DGA

DG A SEQUENCER

3.000E-03EPS-SEQ-FC-DGA

DGA Sequencer FailsTo Operate

1.065E-04EPS-SEQ-CF-DGAB

CCF of DG1A & DG1B Sequencers To Operate

FalseHE-LOOP-A

LOSS OF OFFSITE POWER TO DIVISION A LOADS

EMRALD logic Path Given for External Events simulation1. IE EQ causing LOOP2. Calculation of Peak Ground Acceleration

(PGA) for given EQ3. Evaluate DG availability given EQ (LOOP

SBO yes/no)4. Determine Pipe Failures (Yes/No)

If Yes Run 3D NEUTRINO flooding Simulation

5. Run multiple samples for additional component failure rates (e.g., electrical components), given EQ

6. Call RAVEN/RELAP5-3D given all component failures

7. Log Fuel Damage

9

EMRALD Workflow

NEUTRINO Internal Flooding Model

Switchgear Room 1 – NEUTRINO Flooding Simulation

Components Affected by Flooding 10

to RAVEN/RELAP5-3Dsimulation

EQ-induced SBO• RELAP5-3D nodalization developed/validated for Station Blackout (SBO) transients• Reference reports for Boundary Conditions & Validation:

• US NRC SOARCA Report, NUREG/CR-7110, Vol. 2 • Code-to-code comparison RELAP5-3D/MELCOR

• “Analysis of core damage frequency: Surry, Unit 1 internal events”, NUREG-CR-4550, Vol.3, Rev.1, Pt.1.

• Different SBO scenarios analyzed:– Un-mitigated

• Long-Term SBO (fuel failure in ~14 hrs)• Short-Term SBO (fuel failure in ~2.5 hrs)

– Mitigated• Long-Term SBO (no fuel failure) • Short-Term SBO (fuel failure in ~2.5 hrs)

– Early Failure of MCP considered for all the above cases • MCP leak 182 gpm @ t=+13 min

• Above scenarios bound all possible cases considered by PRA (SAPHIRE) & EMRALD/NEUTRINO

IGPWR – RELAP5-3D modeling• INL RELAP5-3D model for the IGPWR

– 208 volumes / 248 junctions

– 240 HS / 1312 mesh points

Primary System

• RPV, 3 main circulation circuits (SGs, MCPs, HLs and CLs, PRZ)

Secondary side: Steam Lines until MSIV, MFW/AFW inlet

Core configuration:

• 3 hydraulic channels connected with junctions (cross flow simulation) representing 3 different core zones: central, middle and outer core zones [different power]

no BOP

no Containment

12

SG, PRZ, MCP, HL/CL

RPV

Bounding Scenarios – Mitigated LTSBO

EVENT DESCRIPTION

TIME [hh:mm]INL / RELAP5-3D

(SOARCA report / MELCOR)

MCP seal leakage (21 gpm)

Early MCP seal failure (182 gpm)

Initiating event Station blackout – loss of all onsite and offsite AC power 00:00 00:00

Reactor trip, MSIVs close

MCP seals initially leak at 21 gpm/pump (~1 Kg/s)

00:00

(00:00)

00:00

(00:00)

TD-AFW auto initiates at full flow00:01

(00:01)

00:01

(00:01)

RCP seal fail, leaking 182 gpm/pump N/A00:13

(00:13)

First SG SRV opening 00:15

(00:03)

00:15

(00:03)

Operators control TD-AFW to maintain level 00:15

(00:15)

00:15

(00:15)

Void Formation in the UH 01:41 00:27

Operators initiate controlled cooldown of secondary at ~100 F/hr (~55.5 K/hr)

01:30

(01:30)

01:30

(01:30)

UP water level starts to decrease 02:02

(01:57)

00:38

(01:13)

Accumulators begin injecting 02:34

(02:25)

02:15

(02:15)

Vessel water level begins to increase 02:36

(02:30)N/A

Start emergency diesel pump for injection into RCS 03:30

(03:30)

03:30

(03:30)

MCP Seal Leakage

Primary/Secondary Pressures

RPV Level

Kerr Pump Mass Flow

EQ-induced SBO – Bounding Scenarios – Mitigated LTSBO

EVENT DESCRIPTION

TIME [hh:mm]INL / RELAP5-3D

(SOARCA report / MELCOR)

MCP seal leakage (21 gpm)

Early MCP seal failure (182 gpm)

SG cool-down stopped at 120 psig (9.29 MPa) to maintain TD-AFW flow

03:43

(03:35)

03:43

(03:35)

ECST empty. Operator activate a portable, diesel-driven pump (Godwin pump) for supply water to the TD-AFW

~07:35 ~08:44

DC Batteries Exhausted. Operator actions control the secondary pressure at 120 psi and maintain TD-AFW flow

08:00 08:00

Level maintained at the CL elevation with emergency pump N/A 12:38

Core PCT

ECST / AFW Capacity

Primary/Secondary Pressures

Bounding Scenarios• Mitigated STSBO

– Immediate permanent loss of TD-AFW– Recovery actions (e.g., primary side depressurization, Kerr

pump injection) not available before t~2hr– Scenario always ended up in fuel damage no

EMRALD/RELAP5-3D calculations• Mitigated LTSBO & Battery Failure for internal flooding

– Failure of Batteries ( temporary loss of TD-AFW) assumed to happen during first 1 hr from the EQ

– Fuel Failure depending by the recovery time and the MCP leakage rate

– Fuel failures maps help to reduce number of RELAP5-3D calculations

• RELAP5-3D runs executed by EMRALD when we are uncertain about fuel damage status

– e.g., MCP seal leakage 21 gpm, battery failure t=1000 s, 3 hr < recovery time < 3.5 hr

Mitigated STSBO

Mitigated LTSBO + Battery Failure for Internal Flooding

Mitigated LTSBO + Battery Failure for Internal Flooding + Early MCP Seal Failure

Primary/Secondary Pressures RPV Level Core PCT

Automatic Limit Surface Search• Can we do better than this, i.e., automatically

refine the Limit Surface? – identify with more accuracy the boundary

between green (safe) and red (failed) state

– a detailed Limit Surface can avoid the EMRALD/RELAP5-3D on-line calculations

– Manual computer codes runs can be impractical and computationally expensive

• Use of RAVEN code for Automatic Limit Surface Search

– Use of Reduced Order Models (ROM)• Reduce the complexity of the problem• Based on regression/interpolation

techniques• Set of equations are trained to

approximate the original model

Mitigated LTSBO + Battery Failure for Internal Flooding + Early MCP Seal Failure

• Generate a set of training simulations to sample plant response

• Build/update a ROM model• Use ROM to determine an

approximated limit surface• Choose next sample close to the

limit surface• Run simulation for the sampled point

Reachconvergence

Stop

No

Yes

Automatic Limit Surface Search• Several ROM available in RAVEN

– Use of external library Scikit-learn• Open source machine learning library for Python

– Support Vector Machine (SVM) /Classifier ROM used: identify to which category a calculation result belong (safe/failed fuel)

• Train using a set of starting points (RELAP5-3D calculations)

• RBF (exponential) kernel• Good and fast convergence of the surface

• Automatic Limit Surface calculations for different plant scenario (early/not early MCP seal failure, HPI loss, etc.)

• Information contained in the picture is passed via a binary file to EMRALD

SVC + RBF Kernel ROM

Automatic Limit Surface Search

LIMIT SURFACE SEARCH : Mitigated LTSBO + Battery Failure for Internal Flooding + Early MCP Seal Failure

Testing different SVM parameters for convergence

studies

RELAP5-3D/RAVEN Uncertainty Analysis• Quantification of uncertainties on the RELAP5-3D deterministic calculations results

needed (BEPU calculation)• RAVEN code applied to RELAP5-3D also for performing UQ• UQ PIRT for Mitigated-LTSBO

• Important TH phenomena influencing the PCT– NC in primary loop– Secondary Side Mass Inventory loss through SG SRV/ PORV– Primary Side Mass Inventory loss through MCP seal PRZ SRV/PORV– Heat Transfer between primary/secondary system

• Preparing a (partial) list of RELAP5-3D input parameters perturbed by RAVEN code:

– Decay power

– MCP Seal LOCA break area

– Core Pressure losses

– Valves flow areas

– Heat Exchange multiplier

– …

RELAP5-3D/RAVEN Uncertainty Analysis• Selected Input parameters to be perturbed using Monte Carlo

sampler + assigned PDF

• Test the RAVEN code “BasicStatistics” function

– Automatically calculate the basic statistics and matrices (sensitivity, pearson, covariance, etc.)

– identify the most relevant parameters for the selected transient

• Final PCT UQ calculation

– Different approaches possible:

• MC (500-1000 calculations) brute force

• Tolerance Limits (Wilks` formula)

– 59 / 93 / 124 / 153 calculations (first, second, … order statistics)

• Train a meta-model, then perform MC on meta-model

• RAVEN code allows to pursue each of the above approaches

– Exploit INL “FALCON” HPC resources

– Optimized “fast running” RELAP5-3D model: 24hr mission time runs in 1 hr

Sensitivity Parameter

Associated Phenomenon Distribution

Range {±2 σ or

min/max} Power Table Decay Heat Normal ± 7%

Core Pressure Losses

RPV internal circulation Uniform ± 40%

SG PORV/SRV valve flow areas

Critical Flow / Loss of

Secondary Side Mass

Uniform ± 30%

PRZ PORV/SRV

valve flow areas

Critical Flow / Loss of Primary

Side MassUniform ± 30%

MCP seal break area

Critical Flow /Loss of Primary

Side MassUniform ± 20%

SG HX Multiplier

Primary/Secondary Side Heat

ExchangeNormal ± 20%

Initial List of Uncertain Parameters

Sensitivity Parameter Sensitivity Covariance Pearson

Decay Power -1.82E-04 -2.07E+00 -2.94E-01Core Pressure

Losses 1.25E-02 1.43E+02 4.63E-01

SG / PRZ PORV/SRV

valve flow areas4.83E-04 5.49E+00 2.95E-01

MCP seal break area 9.52E-05 1.08E+00 5.80E-02

SG HX Multiplier 1.36E-04 1.55E+00 1.66E-01

Basic Statistics

RELAP5-3D/RAVEN Uncertainty Analysis• 4 parameters selected for final perturbation• MC (1000 runs):

– 1122 K (95%)– E(PCT): 903 K +/- 101

• ….or Wilks` formula applied with LHS: – First/fourth order statistics (59 / 153 runs)

• 59 runs “conservative” value for PCT

LHS sampling PCT resultsCalculation Results (5/Expected/95 Percentile)

RELAP5-3D/RAVEN Uncertainty Analysis• Last Step inform the Limit Surface

Search with the UQ results

– Performing the LSS including epistemic uncertainty

– 6 dimension LSS (4 epistemic, 2 stochastic)

• N-dimensional surface obtained (6-dim)

• Projection of 3 dimensions (Battery Time/Operator action/Core Power)

RELAP5-3D/RAVEN Limit Surface including uncertainty parameters

RELAP5-3D/RAVEN Uncertainty Analysis

• Effects of the epistemic uncertainties restriction of the “safe” area

• More calculations ongoing for achieving better surface resolution fuel failure points imported in EMRLAD

Summary

• Methodology & Tools for LWRS/RISMC External Events analysis defined

• Combined calculations of Structural Mechanics/PRA/Flooding/System Thermal-hydraulic/UQ performed

• Combination of BE advanced tools for the analysis of PWR SBO: MASTODON/EMRALD/NEUTRINO/RAVEN/RELAP5-3D codes

• RAVEN code coupled to RELAP5-3D allowed to inform EMRALD dynamic PRA with BEPU calculations

– Automatic Limit Surface Search

– Computation of Calculation Statistics– Monte Carlo/Tolerance Limits for Uncertainty propagation

24

The National Nuclear Laboratory

25